An Improved Emotion-based Analysis of Arabic Twitter Data using Deep Learning

Abstract

Nowadays everyone is using social media like Twitter, Instagram, Facebook and other social media platforms. Thoughts and feelings about everything could be expressed on these social media platforms. Sentiment and emotion analysis are important tools for analyzing people’s opinions. The lack of using deep learning models in Arabic emotion analysis and the complex structure of the Arabic language encouraged us to explore different word embedding and deep learning models to improve the Arabic emotion analysis accuracy. A combination of Arabic text preprocessing techniques were tested with multiple word embedding, machine learning and deep learning models to categorize the emotion of Arabic tweets into 8 emotions. The AraBERT deep learning model achieved the best accuracy of 75.8% and outperformed other machine learning classifiers in the field of emotion analysis.